28 August 1981, Volume 213, Number 4511 SCI E NCE

Greenhouse Effect
The effective radiating temperature of
the earth, Te, is determined by the need
for infrared emission from the planet to
balance absorbed solar radiation:
Climate Impact of Increasing 'rrR2(1 - A)So = 41TR2cT, (1)
or
Atmospheric Carbon Dioxide Te [So(1 -A)/4or] "
= (2)
J. Hansen, D. Johnson, A. Lacis, S. Lebedeff where R is the radius of the earth, A the
albedo of the earth, S0 the flux of solar
P. Lee, D. Rind, G. Russell radiation, and a the Stefan-Boltzmann
constant. For A 0.3 and So = 1367
-

watts per square meter, this yields

Te 255 K.
-

Atmospheric CO2 increased from 280 The major difficulty in accepting this The mean surface temperature is
to 300 parts per million in 1880 to 335 to theory has been the absence of observed T- 288 K. The excess, Ts - Te, is the
-

340 ppm in 1980 (1, 2), mainly due to warming coincident with the historic greenhouse effect of gases and clouds,
burning of fossil fuels. Deforestation and CO2 increase. In fact, the temperature ir which cause the mean radiating level to
changes in biosphere growth may also the Northern Hemisphere decreased by be above the surface. An estimate of the
greenhouse warming is
Summary. The global temperature rose by 0.20C between the middle 1960's and Te + rH
Ts7 (3)
1980, yielding a warming of 0.4°C in the past century. This temperature increase is where H is the flux-weighted mean alti-
consistent with the calculated greenhouse effect due to measured increases of tude of the emission to space and r is the
atmospheric carbon dioxide. Variations of volcanic aerosols and possibly solar mean temperature gradient (lapse rate)
luminosity appear to be primary causes of observed fluctuations about the mean trend between the surface and H. The earth's
of increasing temperature. It is shown that the anthropogenic carbon dioxide warming troposphere is sufficiently opaque in the
should emerge from the noise level of natural climate variability by the end of the infrared that the purely radiative vertical
century, and there is a high probability of warming in the 1980's. Potential effects on temperature gradient is convectively un-
climate in the 21st century include the creation of drought-prone regions in North stable, giving rise to atmospheric mo-
America and central Asia as part of a shifting of climatic zones, erosion of the West tions that contribute to vertical transport
Antarctic ice sheet with a consequent worldwide rise in sea level, and opening of the of heat and result in r 50 to 6°C per
fabled Northwest Passage. kilometer. The mean lapse rate is less
than the dry adiabatic value because of
latent heat release by condensation as
have contributed, but their net effect is about 0.5°C between 1940 and 1970 (9), a moist air rises and cools and because the
probably limited in magnitude (2, 3). The time of rapid CO2 buildup. In addition, atmospheric motions that transport heat
CO2 abundance is expected to reach 600 recent claims that climate models over- vertically include large-scale atmospher-
ppm in the next century, even if growth estimate the impact of radiative pertur- ic dynamics as well as local convection.
of fossil fuel use is slow (4). bations by an order of magnitude (10, 11) The value of H is -5 km at midlatitudes
Carbon dioxide absorbs in the atmo- have raised the issue of whether the (where r 6.5°C km-') and -6 km in
spheric "window" from 7 to 14 micro- greenhouse effect is well understood. the global mean (r 5.5°C km-1).
meters which transmits thermal radiation We first describe the greenhouse The surface temperature resulting
emitted by the earth's surface and lower mechanism and use a simple model to from flt greenhouse effect is analogous
atmosphere. Increased atmospheric CO2 compare potential radiative perturba- to thE1tfrof water in a leaky bucket
tends to close this window and cause tions of climate. We construct the trend with cmnSWA inflow rate. If the holes in
outgoing radiation to emerge from high- of observed global temperature for the the bucket are reduced slightly in size,
er, colder levels, thus warming the sur- past century and compare this with glob- the water depth and water pressure will
face and lower atmosphere by the so- al climate model computations, provid-
called greenhouse mechanism (5). The ing a check on the ability of the model to The authors are atmospheric physicists at the
most sophisticated models suggest a simulate known climate change. Finally, NASA Institute for Space Studies, Goddard Space
Flight Center, New York 10025. D. Johnson contrib-
mean warming of 2° to 3.5°C for doubling we compute the CO2 warming expected uted to the carbon dioxide research as a participant
of the CO2 concentration from 300 to 600 in the coming century and discuss its in the Summer Institute on Planets and Climate at
the Goddard Institute for Space Studies and Colum-
ppm (6-8). potential implications. bia University.
SCIENCE, VOL. 213, 28 AUGUST 1981 0036-8075/81/0828-0957$01.00/0 Copyright 1981 AAAS 957
increase until the flow rate out of the
holes again equals the inflow rate. Anal-
ogously, if the atmospheric infrared
opacity increases, the temperature of the
surface and atmosphere will increase un-
til the emission of radiation from the
planet again equals the absorbed solar
energy.
The greenhouse theory can be tested
by examination of several planets, which
provide an ensemble of experiments
over a wide range of conditions. The
atmospheric composition of Mars,
Earth, and Venus lead to mean radiating
levels of about 1, 6, and 70 km, and lapse
rates of F 50o, 5.50, and 7°C km-,
respectively. Observed surface tempera-
tures of these planets confirm the exis-
tence and order of magnitude of the
predicted greenhouse effect (Eq. 3). Data
now being collected by spacecraft at
Venus and Mars (12) will permit more
precise analyses of radiative and dynam-
ical mechanisms that affect greenhouse
warming.
One-Dimensional Model
A one-dimensional radiative-convec-
tive (1-D RC) model (5, 13), which com-
putes temperature as a function of alti-
tude, can simulate planetary tempera-
tures more realistically than the zero-
dimensional model of Eq. 1. The
sensitivity of surface temperature in 1-D
RC models to changes in CO2 is similar
to the sensitivity of mean surface tem-
perature in global three-dimensional
models (6-8). This agreement does not
validate the models; it only suggests that
one-dimensional models can simulate the
effect of certain basic mechanisms and
feedbacks. But the agreement does per-
mit useful studies of global mean tem-
perature change with a simple one-di-
mensional model.
The l-D RC model uses a time-march-
ing procedure to compute the vertical
temperature profile from the net radia-
tive and convective energy fluxes:
T(h, t + At) =
T(h, t) + dFr + dFc (4)
cpp dhi dh)
/
where cp is the heat capacity at constant
pressure, p the density of air, h the
altitude, and dF/dh and dFIdh the net
radiative and convective flux diver-
gences. To compute dF/dh the radiative
transfer equation is integrated over all
frequencies, using the temperature pro-
ifie of the previous time step and an
assumed atmospheric composition. The
958
term dFI/dh is the energy transport need-
ed to prevent the temperature gradient
from exceeding a preassigned limit, usu-
ally 6.5°C km-'. This limit parameterizes
effects of vertical mixing and large-scale
dynamics.
The radiative calculations are made by
a method that groups absorption coeffi-
cients by strength for efficiency (14).
Pressure- and temperature-dependent
absorption coefficients are from line-by-
line calculations for H20, C02, 03, N20,
and CH4 (15), including continuum H20
absorption (16). Climatological cloud
cover (50 percent) and aerosol properties
(17) are used, with appropriate fractions
of low (0.3), middle (0.1), and high (0.1)
clouds. Wavelength dependences of
cloud and aerosol properties are ob-
tained from Mie scattering theory (14).
Multiple scattering and overlap of gas-
eous absorption bands are included. Our
computations include the weak CO2
bands at 8 to 12 ,um, but the strong 15-
,um CO2 band, which closes one side of
the 7- to 20-Rm H20 window, causes
- 90 percent of the CO2 warming.
Model Sensitivity
We examine the main processes
known to influence climate model sensi-
tivity by inserting them individually into
the model, as summarized in Table 1.
Model 1 has fixed absolute humidity, a
fixed lapse rate of 6.5°C km-l in the
convective region, fixed cloud altitude,
and no snow/ice albedo feedback or veg-
etation albedo feedback. The increase of
equilibrium surface temperature for dou-
bled atmospheric CO2 is ATs- 1.2°C.
This case is of special interest because it
is the purely radiative-convective result,
with no feedback effects.
Model 2 has fixed relative humidity,
but is otherwise the same as model 1.
The resulting AT, for doubled CO2 is
-
1.9°C. Thus the increasing water vapor
with higher temperature provides a feed-
back factor of -1.6. Fixed relative hu-
midity is clearly more realistic than fixed
absolute humidity, as indicated by physi-
cal arguments (13) and three-dimensional
model results (7, 8). Therefore, we use
fixed relative humidity in the succeeding
experiments and compare models 3 to 6
with model 2.
Model 3 has a moist adiabatic lapse
rate in the convective region rather than
a fixed lapse rate. This causes the equi-
librium surface temperature to be less
sensitive to radiative perturbations, and
AT, - 1.4°C for doubled CO2. The rea-
son is that the lapse rate decreases as
moisture is added to the air, reducing the
temperature difference between the top
of the convective region and the ground
(rH in Eq. 3).
The general circulation of the earth's
atmosphere is driven by solar heating of
the tropical ocean, and resulting evapo-
ration and vertical transport of energy.
The lapse rate is nearly moist adiabatic
at low latitudes and should remain so
after a climate perturbation. Thus use of
a moist adiabatic lapse rate is appropri-
ate for the tropics. But more stable lapse
rates at high latitudes make the surface
temperature much more sensitive to per-
turbations of surface heating (7, 8), and
hence model 3 would underestimate the
sensitivity there. -
Model 4 has the clouds at fixed tem-
perature levels, and thus they move to a
higher altitude as the temperature in-
creases (18). This yields AT, - 2.8°C for
doubled C02, compared to 1.9°C for
fixed cloud altitude. The sensitivity in-
creases because the outgoing thermal
radiation from cloudy regions is defined
by the fixed cloud temperature, requiring
greater adjustment by the ground and
lower atmosphere for outgoing radiation
to balance absorbed solar radiation.
Study of Venus suggests that some
clouds occur at a fixed temperature. The
Venus cloud tops, which are the primary
radiator to space, are at H - 70 km,
where T Te. Analysis of the processes
-
that determine the location of these
clouds and the variety of clouds in the
belts, zones, and polar regions on Jupiter
should be informative. Available evi-
dence suggests that the level of some
terrestrial clouds depends on tempera-
ture while others occur at a fixed alti-
tude. For example, tropical cirrus clouds
moved to a higher altitude in the experi-
ment of Hansen et al. (8) with doubled
C02, but low clouds did not noticeably
change altitude.
Models 5 and 6 illustrate snow/ice and
vegetation albedo feedbacks (19, 20).
Both feedbacks increase model sensitiv-
ity, since increased temperature de-
creases ground albedo and increases ab-
sorption of solar radiation.
Snow, sea ice, and land ice (ice sheets
and glaciers) are all included in snow/ice
albedo feedback. Snow and sea ice re-
spond rapidly to temperature change,
while continental ice sheets require thou-
sands of years to respond. Thus a partial
snow/ice albedo feedback is appropriate
for time scales of 10 to 100 years. The
vegetation albedo feedback was obtained
by comparing today's global vegetation
patterns with reconstruction of the Wis-
consin ice age (20). Uncertainties in the
SCIENCE, VOL. 213
reconstruction, the time scale of vegeta- Table 1. Equilibrium surface temperature increase due to doubled CO2 (from 300 to 600 ppm) in
tion response, and man's potential im- l-D RC models. Model I has no feedbacks affecting the atmosphere's radiative properties.
pact on vegetation prevent reliable as- Feedback factorf specifies the effect of each added process on model sensitivity to doubled
C02; F is the equilibrium thermal flux into the ground if T. is held fixed (infinite heat capacity)
sessment of this feedback, but its esti- when CO2 is doubled. Abbreviations: FRH, fixed relative humidity; FAH, fixed absolute
mated magnitude emphasizes the need to humidity; 6.5LR, 6.5°C km-' limiting lapse rate; MALR, moist adiabatic limiting lapse rate;
monitor global vegetation and surface FCA, fixed cloud altitude, FCT, fixed cloud temperature; SAF, snow/ice albedo feedback; and
albedo. VAF, vegetation albedo feedback. Models 5 and 6 are based onf values from Wang and Stone
(19) and Cess (20), respectively, and AT, of model 2.
Model 4 has our estimate of appropri-
ate model sensitivity. The fixed 6.5°C Model Description AT,
(OC) f F
(W m-2)
km-' lapse rate is a compromise be-
tween expected lower sensitivity at low 1 FAH, 6.5LR, FCA 1.22 1 4.0
latitudes and greater sensitivity at high 2 FRH, 6.5LR, FCA 1.94 1.6 3.9
latitudes. Both cloud temperature and 3 Same as 2, except MALR replaces 6.5LR 1.37 0.7 4.0
snow/ice albedo feedback should be 4 Same as 2, except FCT replaces FCA 2.78 1.4 3.9
5 Same as 2, except SAF included 2.5-2.8 1.3-1.4
partly effective, so for simplicity one is 6 Same as 2, except VAF included -3.5 -1.8
included.
The sensitivity of the climate model
we use is thus ATs 2.8°C for doubled
-

C02, similar to the sensitivity of three-
dimensional climate models (6-8). The
estimated uncertainty is a factor of 2.
This sensitivity (i) refers to perturbations
about today's climate and (ii) does not
include feedback mechanisms effective
only on long time scales, such as changes
of ice sheets or ocean chemistry.
Model Time Dependence
The time dependence of the earth's
surface temperature depends on the heat
capacity of the climate system. Heat
capacity of land areas can be neglected,
since ground is a good insulator. Howev-
er, the upper 100 m of the ocean is
rapidly mixed, so its heat capacity must
be accounted for. The ocean beneath the
mixed layer may also affect surface tem-
perature, if the thermal response time of
the mixed layer is comparable to the time
for exchange of heat with deeper layers.
The great heat capacity of the ocean
and ready exchange of continental and
marine air imply that the global climate
response to perturbations is determined
by the response of the ocean areas.
However, this response is affected by
horizontal atmospheric heat fluxes from
and to the continents. Ready exchange
of energy between the ocean surface and
atmosphere "fixes" the air temperature,
and the ocean in effect removes from the
atmosphere any net heat obtained from
the continents. Thus the horizontal flux
due to a climate perturbation's heating
(or cooling) of the continents adds to the
vertical heat flux into (or out of) the
ocean surface. The net flux into the
ocean surface is therefore larger than it
would be for a 100 percent ocean-cov-
ered planet by the ratio of global area to
ocean area, totaling -5.7 W m-2 for
doubled CO2 rather than -4 W m2. In a
climate model that employs only a
28 AUGUST 1981
mixed-layer ocean, it is equivalent to use
the flux -4 W m-2 with the area-weight-
ed mean land-ocean heat capacity.
The thermal response time of the
ocean mixed layer would be -3 years if
it were not for feedback effects in the
climate system. For example, assume
that the solar flux absorbed by a planet
changes suddenly from Fo orT04 to
F1 = Fo + AF ocT,4, with AF << Fo.
The rate of change of heat in the climate
system is
d(cT)ldt = aT14 - aT4 (5)
where c is heat capacity per unit area.
Since T1 - To << To, the solution is
T - T, = (To - TI)e-t/tthr
where
tthr = c/4aT,3
Thus the planet approaches a new equi-
librium temperature exponentially with
e-folding time tthr. If the heat capacity is
provided by 70 m of water (100 m for
ocean areas) and the effective tempera-
ture is 255 K, tthr is 2.8 years.
This estimate does not account for
climate feedback effects, which can be
analyzed with the 1-D RC model. Table 1
shows that the initial rate of heat storage
in the ocean is independent of feedbacks.
Thus the time needed to reach equilibri-
um for model 4 is larger by the factor
-2.8°C/1.2°C than for model 1, which
excludes feedbacks. The e-folding time
for adjustment of mixed-layer tempera-
ture is therefore -6 years for our best
estimate of model sensitivity to doubled
CO2. This increase in thermal response
time is readily understandable, because
feedbacks come into play only gradually
after some warming occurs.
It would take --50 years to warm up
the thermocline and mixed layer if they
were rapidly mixed, or 250 years for the
entire ocean. Turnover of the deep
ocean, driven by formation of cold bot-
tom water in the North Atlantic and
Antarctic oceans with slow upwelling at
low latitudes, is thought to require 500 to
1000 years (21), suggesting that the deep
ocean does not greatly influence surface
temperature sensitivity. However, there
may be sufficient heat exchange between
the mixed layer and thermocline to delay
full impact of a climate perturbation by a
few decades (6, 22, 23). The primary
mechanism of exchange is nearly hori-
zontal movement of water along surfaces
of constant density (21).
Delay of CO2 warming by the ocean
can be illustrated with a "box diffusion"
model (24), in which heat is stirred in-
(6) stantly through the mixed layer and dif-
fused into the thermocline with diffusion
(7) coefficient k. Observed oceanic penetra-
tion by inert chemical tracers suggests
that k is of order 1 square centimeter per
second (2, 3, 24).
The warming calculated with the one-
dimensional model for the CO2 increase
from 1880 to 1980 (25) is 0.5°C if ocean
heat capacity is neglected (Fig. 1). The
heat capacity of just the mixed layer
reduces this to O.4°C, a direct effect of
the mixed layer's 6-year thermal re-
sponse time. Diffusion into the thermo-
cline further reduces the warming to
0.25°C for k = 1 cm2 sec-', an indirect
effect of the mixed layer's 6-year e-
folding time, which permits substantial
exchange with the thermocline.
The mixed-layer model and thermo-
cline model bracket the likely CO2
warming. The thermocline model is pref-
erable for small climate perturbations
that do not affect ocean mixing. Howev-
er, one effect of warming the ocean sur-
face will be increased vertical stability,
which could reduce ocean warming and
make the surface temperature response
more like that of the mixed-layer case.
Lack of knowledge of ocean processes
959
 1.0 - and global distribution (26). Based on
model calculations, stratospheric aero-
Fig. 1. Dependence of sols that persist for 1 to 3 years after
0.8 CO2 warming on
ocean heat capacity. large volcanic eruptions can cause sub-
Heat is rapidly mixed stantial cooling of surface air (Fig. 2).
in the upper 100 m of The cooling depends on the assumption
0.6 the ocean and dif- that the particles do not exceed a few
0 fused to 1000 m with tenths of a micrometer in size, so they do
0L diffusion coefficient k.
I- 0.4 The CO2 abundance, not cause greenhouse warming by block-
from (25), is 293 ppm ing terrestrial radiation, but this condi-
in 1880, 335 ppm in tion is probably ensured by rapid gravita-
1980, and 373 ppm in tional settling of larger particles. Tempo-
0.2 2000. Climate model ral variability of stratospheric aerosols
equilibrium sensitiv-
ity is 2.8°C for dou- due to volcanic eruptions appears to
0 - bled CO2. have been responsible for a large part of
1880 1900 1920 1940 1960 1980 2000 the observed climate change during the
Date past century (27-30), as shown below.
The impact of tropospheric aerosols
on climate is uncertain in sense and
primarily introduces uncertainties about A 1 percent increase of solar luminos- magnitude due to their range of composi-
the time dependence of the global CO2 ity would warm the earth 1.6°C at equi- tion, including absorbing material such
warming. The full impact of the warming librium (Fig. 2) on the basis of model 4, as carbon and high-albedo material such
may be delayed several decades, but which we employ for all radiative pertur- as sulfuric acid, and their heterogeneous
since man-made increases in atmospher- bations to provide a uniform compari- spatial distribution. Although man-made
ic CO2 are expected to persist for centu- son. Since the effect is linear for small tropospheric aerosols are obvious near
ries (1, 2, 6), the warming will eventually changes of solar luminosity, a change of their source, aerosol opacity does not
occur. 0.3 percent would modify the equilibri- appear to have increased much in remote
um global mean temperature by 0.5°C, regions (31). Since the climate impact of
which is as large as the equilibrium anthropogenic aerosols is also reduced
Radiative Climate Perturbations warming for the cumulative increase of by the opposing effects of absorbing and
atmospheric CO2 from 1880 to 1980. So- high-albedo materials, it is possible that
Identification of the CO2 warming in lar luminosity variations of a few tenths they have not had a primary effect on
observed climate depends on the magni- of 1 percent could not be reliably mea- global temperature. However, global
tude of climate variability due to other sured with the techniques available dur- monitoring of aerosol properties is need-
factors. Most suspected causes of global ing the past century, and thus are a ed for conclusive analysis.
climate change are radiative perturba- possible cause of part of the climate Ground albedo alterations associated
tions, which can be compared to identify variability in that period. with changing patterns of vegetation
those capable of counteracting or rein- Atmospheric aerosol effects depend coverage have been suggested as a cause
forcing the CO2 warming. on aerosol composition, size, altitude, of global climate variations on time
scales of decades to centuries (32). A
global surface albedo change of 0.015,
Potential radiative perturbations of climate equivalent to a change of 0.05 over land
4 Warming 0 Cooling areas, would affect global temperature
by 1.3°C. Since this is a 25 percent
change in mean continental ground albe-
3 do, it seems unlikely that ground albedo
0 variations have been the primary cause
0
-1.9
of recent global temperature trends.
_- 2 However, global monitoring of ground
4
-1.3 -1 .4 albedo is needed to permit definitive
assessment of its role in climate variabili-
ty.
High and low clouds have opposite
0 effects on surface temperature (Fig. 2),
high clouds having a greenhouse effect
C02 Strat, Tropo. Low High CH4
(300 ppm - aerosols aerosols clouds clouds (1.6 ppm while low clouds cool the surface (14,
600 ppm) H2804 Soot (+2% of (+2% of -3.2 ppm)
33). However, the nature and causes of
(|&ru+0.2) (&rT+0.02) globe) globe) variability of cloud cover, optical thick-
Solar Tropo. Land Middle IN20 CCI2F2 & ness, and altitude distribution are not
luminosity aerosols albedo clouds (0.228 ppm CCI3F well known, nor is it known how to
(+1%) H2804 (+0.05) (+2% of -0.56 ppm)(0-2 ppb model reliably cloud feedbacks that may
(Aru+0. 1) globe) each) occur in response to climate perturba-
tions. Progress may be made after accu-
Fig. 2. Surface temperature effect of various global radiative perturbations, based on the l-D rate cloud climatology is obtained from
RC model 4 (Table 1). Aerosols have the physical properties specified by (17). Dependence of
AT on aerosol size, composition, altitude, and optical thickness is illustrated by (26). The AT for global observations, including seasonal
stratospheric aerosols is representative of a very large volcanic eruption. and interannual cloud variations. In the
960 SCIENCE, VOL. 213
meantime, some limits are implicitly
placed on global cloud feedback by em-
pirical tests of the climate system's sen-
sitivity to radiative perturbations, as dis-
cussed below.
Trace gases that absorb in the infrared
can warm the earth if their abundance
increases (5, 34). The abundance of
chlorofluorocarbons (Freons) increased
from a negligible amount a few decades
ago to 0.3 part per billion for CC12F2 and
0.2 ppb for CC13F (35), with an equilibri-
um greenhouse warming of 0.06°C. -
Recent measurement of a 0.2 percent per
year increase of N20 suggests a cumula-
tive increase to date of 17 ppb (36), with
an equilibrium warming of - 0.03°C.
Tentative indications of a 2 percent per
year increase in CH4 imply an equilibri-
um warming < 0.1°(C for the CH4 in-
crease to date (37). No major trend of 03
abundance has been observed, although
it has been argued that continued in-
crease of Freons will reduce 03 amounts
(38). The net impact of measured trace
gases has thus been an equilibrium
warming of 0. 1°C or slightly larger. This
does not greatly alter analyses of tem-
perature change over the past century,
but trace gases will significantly enhance
future greenhouse warming if recent
growth rates are maintained.
We conclude that study of global cli-
mate change on time scales of decades
and centuries must consider variability
of stratospheric aerosols and solar lumi-
nosity, in addition to CO2 and trace
gases. Tropospheric aerosols and ground
albedo are potentially significant, but
require better observations. Cloud vari-
ability will continue to cause uncertainty
until accurate monitoring of global cloud
properties provides a basis for realistic
modeling of cloud feedback effects; how-
ever, global feedback is implicitly
checked by comparison of climate model
sensitivity to empirical climate varia-
tions, as done below.
Observed Temperature Trends
Data archives (39) contain surface air
temperatures of several hundred stations
for the last century. Problems in obtain-
ing a global temperature history are due
to the uneven station distribution (40),
with the Southern Hemisphere and
ocean areas poorly represented, and the
smaller number of stations for earlier
times.
We combined these temperature rec-
ords with a method designed to extract
mean temperature trends. The globe was
divided by grids with a spacing not larger
than the correlation distance for primary
28 AUGUST 1981
dynamical transports (41), but large
enough that most boxes contained one or
more stations. The results shown were
obtained with 40 equal-area boxes in
each hemisphere, but the conclusions
are not sensitive to the exact spacing.
Temperature trends for stations within a
box were combined successively:
T,n(t) (n* -
l)T1,n + Tn -
Tn +
to obtain a single trend for each box,
where the bar indicates a mean for the
years in which there are records for both
Tn and the cumulative T1,n and n*(t) is
the number of stations in Tj,m(t). Trends
for boxes in a latitude zone were com-
bined with each box weighted equally,
and the global trend was obtained by
area-weighting the trends for all latitude
zones. A meaningful result begins in the
1880's, since thereafter continuous rec-
ords exist for at least two widely separat-
ed longitudes in seven of the eight lati-
tude zones (continuous Antarctic tem-
peratures begin in the 1950's). Results
are least reliable for 1880 to 1900; by
1900, continuous records exist for more
than half of the 80 boxes.
The temperature trends in Fig. 3 are
smoothed with a 5-year running mean to Model Verification
make the trends readily visible. Part of
the noise in the unsmoothed data results Natural radiative perturbations of the
from unpredictable weather fluctuations, earth's climate, such as those due to
which affect even 1-year means (42). aerosols produced by large volcanic
0.4
0.2
0
o -0.2
-
-0.4 f
Fig. 3. Observed surface air 0.2
temperature trends for three
latitude bands and the entire
globe. Temperature scales for
low latitudes and global mean -0.2
are on the right.
None of our conclusions depends on the
nature of the smoothing.
Northern latitudes warmed - 0.8UC
between the 1880's and 1940, then
cooled - 0.5C between 1940 and 1970,
m agreement with other analyses (9, 43).
Low latitudes warmed - 0.3°C between
1880 and 1930, with little change thereaf-
ter. Southern latitudes warmed - 0.4°C
TI,n in the past century; results agree with a
prior analysis for the late 1950's to mid-
(8) dle 1970's (44). The global mean tem-
perature increased 0.5°C between
-
1885 and 1940, with slight cooling there-
after.
A remarkable conclusion from Fig. 3 is
that the global temperature is almost as
high today as it was in 1940. The com-
mon misconception that the world is
cooling is based on Northern Hemi-
sphere experience to 1970.
Another conclusion is that global sur-
face air temperature rose - 0.4OC in the
past century, roughly consistent with
calculated CO2 warming. The time his-
tory of the warming obviously does not
follow the course of the CO2 increase
(Fig. 1), indicating that other factors
miust affect global mean temperature.
Obserived temperature (5-year running mean)
0.2
0
owlatituides
-0.2
0outhern latitudes
zX~(3.608-900S)0.
.JX J\\ Al 0.2
Date
961
AS=O.1 A Ft* -2.4 Net-+2.5
t I
I1 Atmos.
-0.3 + Surface
Ts = 287.5
(a) Imrmediate response
Fig. 4. Change of fluxes (watts per square meter) in the 1-D RC model when atmospheric CO2 is doubled (from 300 to 600 ppm). Symbols: AS,
change in solar radiation absorbed by the atmosphere and surface; AF t , change in outward thermal radiation at top of the atmosphere. The wavy
line represents convective flux; other fluxes are radiative.
eruptions, permit a valuable test of tnod-
el sensitivity. Previous study of the best-
documented large volcanic eruption,
Mount Agung in 1963, showed that tropi-
cal tropospheric and stratospheric tem-
perature changes computed with a one-
dimensional climate model were of the
same sign and order of magnitude as
observed changes (45). It was assumed
that horizontal heat exchange with high-
er latitudes was not altered by the radia-
tive perturbation.
We reexamined the Mount Agung case
for comparison with the present global
temperature record, using our model
with sensitivity 2.8°C. The model,
-
with a maximum global mean aerosol
increase in the optical depth AT = 0.12
(45), yields a maximum global cooling of
0.2°C when only the mixed-layer heat
capacity is included and 0. 1C when heat
exchange with the deeper ocean is in-
cluded with k = 1 cm2 sec-1. Observa-
tions suggest a cooling of this magnitude
with the expected time lag of 1 to 2
years. Noise or unexplained variability
in the observations prevents more defini-
tive conclusions, but similar cooling is
indicated by statistical studies of tem-
perature trends following other large vol-
canic eruptions (46).
A primary lesson from the Mount
Agung test is the damping of temperature
change by the mixed layer's heat capaci-
ty, without which the cooling would
have exceeded 1. 1C (Fig. 2). The effect
can be understood from the time con-
stant of the perturbation and thermal
response time of the mixed layer:
AT {1 exp[(- 1 year)/(6 years)]} x
- serve energy. There is a small increase in
1.1°C 0.17C, for the case k = 0. This
large reduction of the clitnate response
occurs for a perturbation that (unlike
C02) is present for a time shorter than
the thermal response time of the ocean
surface.
Phenomena that alter the regional radi-
ation balance provide another model
test. Idso (11) found a consistent "em-
pirical response function" for several
such phenomena, which was 0. 17C per
962
AS*0.1 AFtu-3.8
+0.4
t
-0.8
t
-30
-0.3 -3.1 8 + 1.
;0 3 .7 ,7g,
, i 7.2
Ts = 287.5
(b) A few months later
watt per square meter in midcontinent
and was half as large on the coast. This
response must depend on the rate of
mixing of marine and continental air,
since the phenomena occur on time
scales less than the thermal relaxation
time of the ocean surface. Thus, as one
test of horizontal atmospheric trans-
ports, we read from our three-dimen-
sional climate model (8) the quantities
(solar insolation and temperature) that
form Idso's empirical response function
for seasonal change of insolation. Re-
sults ranged from 0.2°C W' m2 in mid-
continent, and about half that on the
coast, to a value an order of magnitude
smaller over the ocean, in agreement
with the empirical response (11).
To relate these empirical tests to the
CO2 greenhouse effect, we illustrate the
flux changes in the I-D RC model when
CO2 is doubled. For simplicity we con-
sider an instantaneous doubling of C02,
and hence the time dependence of the
response does not represent the transient
response to a steady change in CO2. The
immediate response to the doubling in-
cludes (Fig. 4a): (i) reduced emission to
space (- 2.4 W m-2), because added CO2
absorption raises the mean altitude of
emission to a higher, colder level; (ii)
increased flux from atmosphere to
ground (+ 1.1 W m2); and (iii) increased
stratospheric cooling but decreased tro-
pospheric cooling. The radiative warm-
ing of the troposphere decreases the
convective" flux (latent. and sensible
heat) from the ground by 3.5 W m72 as a
consequence of the requirement to con-
absorption of near-infrared radiation, the
atmosphere gaining energy (+ 0.4 W
mf2) and the ground losing energy (- 0.3
W m72). The net effect is thus an energy
gain for the planet (+ 2.5 W m-2) with
heating of the ground (+ 4.3 W m-2) and
cooling of the (upper) atmosphere (- 1.8
W m72). These flux changes are indepen-
dent offeedbacks and are not sensitive to
the critical lapse rate.
A few months after the CO2 doubling
Net*+3.9 AS-0.l AF' -0.1 NetO0
Atmos.
t t
+1.9 -1.4 +1.5 Atmos.
--U-\--- - -1-T- E
Surface -1.8 0 +171 Surface
1+I6i6 17.
Ts = 290.3
(c) Many years later
(Fig. 4b) the stratospheric temperature
has cooled by - 5°C. Neither the ocean
nor the troposphere, which is convec-
tively coupled to the surface, have re-
sponded yet. The planet radiates 3.8 W
m72 less energy to space than in the
comparison case with 300 ppm CO2,
because of the cooler stratosphere and
greater altitude of emission from the
troposphere. The energy gained by the
earth at this time is being used to warm
the ocean.
Years later (Fig. 4c) the surface tem-
perature has increased 2.8C. Almost
half the increase (1.2°C) is the direct CO2
greenhouse effect. The remainder is due
to feedbacks, of which 1.0°C is the well-
established H20 greenhouse effect.
The greenhouse process represented
in Fig. 4 is simply the "leaky bucket"
phenomenon. The increased infrared
opacity causes an immediate decrease of
thermal radiation from the planet, thus
forcing the temperature to rise until ener-
gy balance is restored. Temporal varia-
tions of the fluxes and temperatures are
due to the response times of the atmo-
sphere and surface.
Surface warming of - 3°C for doubled
CO2 is the status after energy balance
has been restored. This contrasts with
the Agung case and the cases considered
by Idso (11), which are all nonequilibri-
um situations.
The test of the greenhouse theory pro-
vided by the extremes of equilibrium
climates on the planets and short-term
radiative perturbations is reassuring, but
inadequate. A crucial intermediate test is
climate change on time scales from a few
years to a century.
Model versus Observations for the
Past Century
Simulations of global temperature
change should begin with the known
forcings: variations of CO2 and volcanic
aerosols. Solar luminosity variations,
which constitute another likely mecha-
SCIENCE, VOL. 213
nism, are unknown, but there are
hypotheses consistent with observation-
al constraints that variations not exceed
a few tenths of 1 percent.
We developed an empirical equation
that fits the heat flux into the earth's
surface calculated with the l-D RC cli-
mate model (model 4):
F(t) = 0.018Ap/(l + 0.0022Ap)06-
17AT - 1.5(AT)2 + 220AS/5O -
1.5AT + 0.033 (Al)2 - 1.04 X
10ApAT + 0.29ATAT (9)
where F(t) is in watts per square meter, p
is the amount of CO2 in parts per million
above an "equilibrium" value (293
ppm), AS is the difference between solar
luminosity and an equilibrium value So,
AT iS the optical depth of stratospheric
aerosols above a background amount,
and AT is the difference between current
surface temperature and the equilibrium
value for Ap = AS = AT = 0. Equation
9 fits the one-dimensional model results
to better than 1 percent for 0 s Ap
< 1200 ppm, 0.98 c AS/SO c 1.02, and
AT s 0.5. For the mixed-layer ocean
model T,(t) follows from dTI/dt = F(t)l
co, where co is the heat capacity of the
ocean mixed layer per unit area. If the
true mixed-layer depth is used to obtain
co, F(t) must be multiplied by 1/0.7, the
ratio of global area to ocean area. Diffu-
sion of heat into the deeper ocean can
then also be included by means of the
diffusion equation with T. as its upper
boundary condition.
The CO2 abundance increased from
293 ppm in 1880 to 335 ppm in 1980 (25),
based on recent accurate observations,
earlier less accurate observations, and
carbon cycle modeling. The error for
1880 probably does not exceed 10 ppm
(1, 2).
Volcanic aerosol radiative forcing can
be obtained from Lamb's (27) dust veil
index (DVI), which is based mainly on
atmospheric transmission measurements
after 1880. We convert DVI to optical
depth by taking Mount Agung (DVI =
800) to have the maximum AT = 0.12.
The aerosol optical depth histories of
Mitchell (47) and Pollack et al. (29), the
latter based solely on transmission mea-
surements, are similar to Lamb's. We
use aerosol microphysical properties
from (45). The error in volcanic aerosol
radiative forcing probably does not ex-
ceed a factor of 2.
Solar variability is highly conjectural,
so we first study CO2 and volcanic aero-
sol forcings and then add solar varia-
tions. We examine the hypothesis of
Hoyt (48) that the ratio, r, of umbra to
penumbra areas in sunspots is pro-
28 AUGUST 1981
portional to solar luminosity: AS/
So = fir - ro). Hoyt's rationale is that
the penumbra, with a weaker magnetic
field than the umbra, is destroyed more
readily by an increase of convective flux
from below. We take f = 0.03, which
implies a peak-to-peak amplitude of
0.4 percent for AS/S0 in the past cen-
-
tury, or an amplitude of - 0.2 percent
for the mean trend line. Taking So as the
mean for 1880 to 1976 yields ro = 0.2.
The resulting AS/S0 has no observational
corroboration, but serves as an example
of solar variability of a plausible magni-
tude.
Radiative forcing by CO2 plus volca-
noes and forcing by CO2 plus volcanoes
plus the sun both yield a temperature
trend with a strong similarity to the ob-
served trend of the past century (Fig. 5),
which we quantify below. If only the
heat capacity of the mixed layer is in-
cluded, the amplitude of the computed
temperature variations is larger than ob-
served. However, mixing of heat into the
deeper ocean with k = 1 cm2 sec-1
brings both calculated trends into rough
agreement with observations.
The main uncertainties in the climate
model-that is, its "tuning knobs"-are
(i) the equilibrium sensitivity and (ii) the
rate of heat exchange with the ocean
beneath the mixed layer. The general
correlation of radiative forcings with
Ocean model: mixed layer only
0.2
0
-0.2
0
0
00
-0.2
-0.4
-0.6 I
1 880 1920 1960
Date
Fig. 5. Global temperature trend obtained from climate model with sensitivity 2.8°C for doubled
CO2. The results in (a) are based on a 100-m mixed-layer ocean for heat capacity; those in (b)
include diffusion of heat into the thermocline to 1000 m. The forcings by C02, volcanoes, and
the sun are based on Broecker (25), Lamb (27), and Hoyt (48). Mean ATis zero for observations
and model.
global temperatures suggests that model
uncertainties be constrained by requiring
agreement with the observed tempera-
ture trend.
Therefore, we examined a range of
model sensitivities, choosing a diffusion
coefficient for each to minimize the re-
sidual yariance between computed and
observed temperature trends. Equilibri-
um sensitivities of 1.4°, 2.80, and 5.6°C
required k = 0, 1.2, and 2.2 cm2 sec-',
respectively. All models with sensitiv-
ities of 1.4° to 5.6°C provide a good fit to
the observations. The smallest accept-
able sensitivity is - 1.4°C, because it
requires zero heat exchange with the
deeper ocean. Sensitivities much higher
than 5.60C would require greater heat
exchange with the deep ocean than is
believed to be realistic (21, 22).
Radiative forcing by CO2 plus volca-
noes accounts for 75 percent of the vari-
ance in the 5-year smoothed global tem-
perature, with correlation coefficient 0.9.
The hypothesized solar luminosity varia-
tion (48) improves the fit, as a conse-
quence of the luminosity peaking in the
1930's and declining into the 1970's,
leaving a residual variance of only 10
percent. The improved fit provided by
Hoyt's solar variability represents a pos-
teriori selection, since other hypothe-
sized solar variations that we examined
[for instance (49)] degrade the fit. This
Mixed layer and thermocilne
(k= 1 cm2 eec-1)
1900 1940 1980
963
 Table 2. Energy supplied and CO2 released by fuels.
Energy
supplied
Fuel Fuel in 1980*
in 1980*
(oil = 1)
(1019J) (%)
Oil 12 40
Coal 7 24
Gas 5 16
Oil shale, tar sands, 0 0
heavy oil
Nuclear, solar, wood, 6 20
hydroelectric
Total 30 100
*Based on late 1970's. tReservoir estimates assume that half the coal above 3000 feet can be recovered
and that oil recovery rates will increase from 25 to 30 percent to 40 percent. Estimate for unconventional
fossil fuels may be low if techniques are developed for economic extraction of "synthetic oil" from deposits
that are deep or of marginal energy content. It is assumed that the airborne fraction of released CO2 is fixed.
evidence is too weak to support any
specific solar variability.
The general agreement between mod-
eled and observed temperature trends
strongly suggests that CO2 and volcanic
aerosols are responsible for much of the
global temperature variation in the past
century. Key consequences are: (i) em-
pirical evidence that much of the global
climate variability on time scales of de-
cades to centuries is deterministic and
(ii) improved confidence in the ability of
models to predict future CO2 climate
effects.
Projections into the 21st Century
Prediction of the climate effect of CO2
requires projections of the amount of
atmospheric C02, which we specify by
(i) the energy growth rate and (ii) the
fossil fuel proportion of energy use. We
neglect other possible variables, such as
changes in the amount of biomass or the
fraction of released CO2 talken up by the
ocean.
Energy growth has been 4 to 5 percent
per year in the past century, but increas-
ing costs will constrain future growth (1,
4). Thus we consider fast growth (- 3
percent per year, specifically 4 percent
per year in 1980 to 2020, 3 percent per
year in 2020 to 2060, and 2 percent per
year in 2060 to 2100), slow growth (half
of fast growth), and no growth as repre-
sentative energy growth rates.
Fossil fuel use will be limited by avail-
able resources (Table 2). Full use of oil
and gas will increase CO2 abundance by
< 50 percent of the preindustrial amount.
Oil and gas depletion are near the 25
percent level, at which use of a resource
normally begins to be limited by supply
and demand forces (4). But coal, only 2 to
3 percent depleted, will not be so con-
strained for several decades.
964
Airborne
C02 C02
release CO2
added added
perunit
per unit 1980*in through
1980
(ppm)
(%) (ppm)
1 50 0.7 11
5/4 35 0.5 26
3/4 15 0.2 5 50
7/4 0 0 0 100
0 0 0 0
100 1.4 42
The key fuel choice is between coal
and alternatives that do not increase
atmospheric CO2. We examine a synfuel
option in which coal-derived synthetic
fuels replace oil and gas as the latter are
depleted, and a nuclear/renewable re-
sources option in which the replacement
fuels do not increase C02. We also ex-
amine a coal phaseout scenario: after a
specific date coal and synfuel use are
held constant for 20 years and then
phased out linearly over 20 years.
Projected global warming for fast
growth is 3° to 4.5°C at the end of the
next century, depending on the propor-
tion of depleted oil and gas replaced by
synfuels (Fig. 6). Slow growth, with de-
pleted oil and gas replaced equally by
synfuels and nonfossil fuels, reduces the
warming to 2.5°C. The warming is
-
only slightly more than 1°C for either (i)
no energy growth, with depleted oil and
gas replaced by nonfossil fuels, or (ii)
slow energy growth, with coal and syn-
fuels phased out beginning in 2000.
Other climate forcings may counteract
or reinforce CO2 warming. A decrease of
solar luminosity from 1980 to 2100 by 0.6
percent per century, large compared to
measured variations, would decrease the
warming 0.7°C. Thus CO2 growth as
large as in the slow-growth scenario
would overwhelm the effect of likely
solar variability. The same is true of
other radiative perturbations; for in-
stance, volcanic aerosols may slow the
rise in temperature, but even an optical
thickness of 0.1 maintained for 120 years
would reduce the warming by less than
1.00C.
When should the CO2 warming rise
out of the noise level of natural climate
variability? An estimate can be obtained
by comparing the predicted warming to
the standard deviation, a, of the ob-
served global temperature trend of the
past century (50). The standard devi-
ation, which increases from 0. 1°C for 10-
year intervals to 0.20C for the full centu-
Potential ry, is the total variability of global tem-
airbomn
virgin perature; it thus includes variations due
virst to any known radiative forcing, other
variations of the true global temperature
(ppm) due to unidentified causes, and noise due
70 to imperfect measurement of the global
1000 temperature. Thus if To is the current 5-
year smoothed global temperature, the 5-
year smoothed global temperature in 10
0 years should be in the range To ± 0. 1°C
with probability - 70 percent, judging
1220 only from variability in the past century.
The predicted CO2 warming rises out
of the la noise level in the 1980's and the
2a level in the 1990's (Fig. 7). This is
independent of the climate model's equi-
librium sensitivity for the range of likely
values, 1.4° to 5.6°C. Furthermore, it
does not depend on the scenario for
atmospheric CO2 growth, because the
amounts of CO2 do not differ substantial-
ly until after year 2000. Volcanic erup-
tions of the size of Krakatoa or Agung
may slow the warming, but barring an
unusual coincidence of eruptions, the
delay will not exceed several years.
Nominal confidence in the CO2 theory
will reach 85 percent when the tem-
-
perature rises through the lr level and
-
98 percent when it exceeds 2u. How-
ever, a portion of a may be accounted
for in the future from accurate knowl-
edge of some radiative forcings and more
precise knowledge of global tempera-
ture. We conclude that CO2 warming
should rise above the noise level of natu-
ral climate variability in this century.
Potential Consequences of
Global Warming
Practical implications of CO2 warming
can only be crudely estimated, based on
climate models and study of past cli-
mate. Models do not yet accurately sim-
ulate many parts of the climate system,
especially the ocean, clouds, polar sea
ice, and ice sheets. Evidence from past
climate is also limited, since the few
recent warm periods were not as ex-
treme as the warming projected to ac-
company full use of fossil fuels, and the
climate forcings and rate of climate
change may have been different. Howev-
er, if checked against our understanding
of the physical processes and used with
caution, the models and data on past
climate provide useful indications of pos-
sible future climate effects (51).
Paleoclimatic evidence suggests that
surface warming at high latitudes will be
two to five times the global mean warm-
ing (52-55). Climate models predict the
SCIENCE, VOL. 213
larger sensitivity at high latitudes and Danger of rapid sea level rise is posed sheets. A sea level rise of 5 m would
trace it to snow/ice albedo feedback and by the West Antarctic ice sheet, which, flood 25 percent of Louisiana and Flori-
greater atmospheric stability, which unlike the land-based Greenland and da, 10 percent of New Jersey, and many
magnifies the warming of near-surface East Antarctic ice sheets, is grounded other lowlands throughout the world.
layers (6-8). Since these mechanisms below sea level, making it vulnerable to Climate models (7, 8) indicate that
will operate even with the expected ra- rapid disintegration and melting in case - 2°C global warming is needed to cause
pidity of CO2 warming, it can be antici- of general warming (55). The summer - 5°C warming at the West Antarctic ice
pated that average high-latitude warming temperature in its vicinity is about -5°C. sheet. A 2°C global warming is exceeded
will be a few times greater than the If this temperature rises - 5°C, deglacia- in the 21st century in all the CO2 scenari-
global mean effect. tion could be rapid, requiring a century os we considered, except no growth and
Climate models indicate that large re- or less and causing a sea level rise of 5 to coal phaseout.
gional climate variations will accompany 6 m (55). If the West Antarctic ice sheet Floating polar sea ice responds rapidly
global warming. Such shifting of climatic melts on such a time scale, it will tempo- to climate change. The 5° to 10°C warm-
patterns has great practical significance, rarily overwhelm any sea level change ing expected at high northern latitudes
because the precipitation patterns deter- due to growth or decay of land-based ice for doubled CO2 should open the North-
mine the locations of deserts, fertile ar-
eas, and marginal lands. A major region-
al change in the doubled CO2 experiment
with our three-dimensional model (6, 8)
was the creation of hot, dry conditions in 4
much of the western two-thirds of the
United States and Canada and in large
parts of central Asia. The hot, dry sum-
mer of 1980 may be typical of the United 0
2)
States in the next century if the model Fig. 6. Projections 3
results are correct. However, the model of global tempera-
ture. The diffusion
shows that many other places, especially coefficient beneath
coastal areas, are wetter with doubled the ocean mixed
CO2. layer is 1.2 cm2 I-
Reconstructions of regional climate sec', as required -0 T 'a
patterns in the altithermal (53, 54) show for best fit of the .Yco
model and observa-
some similarity to these model results. tions for the period t-

The United States was drier than today 1880 to 1978. Esti- X
during that warm period, but most re- mated global mean 1
gions were wetter than at present. For warming in earlier . ,E
warm periods is in-
example, the climate in much of North dicated
Africa and the Middle East was more
on the right. i -
favorable for agriculture 8000 to 4000
years ago, at the time civilization 0
dawned in that region.
Beneficial effects of CO2 warming will
include increased length of the growing 1950 2000 2050 2100
season. It is not obvious whether the Date
world will be more or less able to feed its
population. Major modifications of re-
gional climate patterns will require ef- 0.8
forts to readjust land use and crop char- C02 warming versus noise level of natural climate varlabilitv
acteristics and may cause large-scale hu-
man dislocations. Improved global cli- 0.6 F Model sensitivity
1 - 19aty -apiit
C&
- - - AT = 5.60C
A~~
mate models, reconstructions of past AT = 2.80C
climate, and detailed analyses are need- 0.4 _ AT = 1.40C
ed before one can predict whether the 0

net long-term impact will be beneficial or 0

A ..~~~~~--
detrimental. .4
0.2 -2a of natural variability,
Melting of the world's ice sheets is lco of natural variability NNb
another possible effect CO2 warming.
of -~~~~~~~~
0 ....I . ..l

If they melted entirely, sea level would

rise 70 m. However, their natural re-
-
Observed *-..--
sponse time is thousands of years, and it -0.2 temperature
is not certain whether CO2 warming will
cause the ice sheets to shrink or grow. 1950 1960 1970 1980 1990 2000 2010 2020
For example, if the ocean warms but the Date
air above the ice sheets remains below Fig. 7. Comparison of
freezing, the effect could be increased temperature projected C02 warming to standard deviation (o) of observed global
and to 2r. The standard deviation was computed for the observed global
snowfall, net ice sheet growth, and thus temperatures in Fig. 3. Carbon dioxide change is from the slow-growth scenario. The effect of
lowering of sea level. other trace gases is not included.
28 AUGUST 1981
west and Northeast passages along the
borders of the American and Eurasian
continents. Preliminary experiments
with sea ice models (56) suggest that all
the sea ice may melt in summer, but part
of it would refreeze in winter. Even a
partially ice-free Arctic will modify
neighboring continental climates.
Discussion
The global warming projected for the
next century is of almost unprecedented
magnitude. On the basis of our model
calculations, we estimate it to be
-
2.5°C for a scenario with slow energy
growth and a mixture of nonfossil and
fossil fuels. This would exceed the tem-
perature during the altithermal (6000
years ago) and the previous (Eemian)
interglacial period 125,000 years ago
(53), and would approach the warmth of
the Mesozoic, the age of dinosaurs.
Many caveats must accompany the
projected climate effects. First, the in-
crease of atmospheric CO2 depends on
the assumed energy growth rate, the
proportion of energy derived from fossil
fuels, and the assumption that about 50
percent of anthropogenic CO2 emissions
will remain airborne. Second, the pre-
dicted global warming for a given CO2
increase is based on rudimentary abili-
ties to model a complex climate system
with many nonlinear processes. Tests of
model sensitivity, ranging from the equi-
librium climates on the planets to pertur-
bations of the earth's climate, are en-
couraging, but more tests are needed.
Third, only crude estimates exist for
regional climate effects.
More observations and theoretical
work are needed to permit firm identifi-
cation of the CO2 warming and reliable
prediction of larger climate effects far-
ther in the future. It is necessary to
monitor primary global radiative forc-
ings: solar luminosity, cloud properties,
aerosol properties, ground albedo, and
trace gases. Exciting capabilities are
within reach. For example, the NASA
Solar Maximum Mission is monitoring
solar output with a relative accuracy of
-
0.01 percent (57). Studies of certain
components of the climate system are
needed, especially heat storage and
transport by the oceans and ice sheet
dynamics. These studies will require
global monitoring and local measure-
ments of processes, guided by theoreti-
cal studies. Climate models must be de-
veloped to reliably simulate regional cli-
mate, including the transient response
966
(58) to gradually increasing CO2 amount.
Political and economic forces affecting
energy use and fuel choice make it un-
likely that the CO2 issue will have a
major impact on energy policies until
convincing observations of the global
warming are in hand. In light of historical
evidence that it takes several decades to
complete a major change in fuel use, this
makes large climate change almost inev-
itable. However, the degree of warming
will depend strongly on the energy
growth rate and choice of fuels for the
next century. Thus, CO2 effects on cli-
mate may make full exploitation of coal
resources undesirable. An appropriate
strategy may be to encourage energy
conservation and develop alternative en-
ergy sources, while using fossil fuels as
necessary during the next few decades.
The climate change induced by anthro-
pogenic release of CO2 is likely to be the
most fascinating global geophysical ex-
periment that man will ever conduct.
The scientific task is to help determine
the nature of future climatic effects as
early as possible. The required efforts in
global observations and climate analysis
are challenging, but the benefits from
improved understanding of climate will
surely warrant the work invested.
References and Notes
1. National Academy of Sciences, Energy and
Climate (Washington, D.C., 1977).
2. U. Siegenthaler and H. Oeschger, Science 199,
388 (1978).
3. W. S. Broecker, T. Takahashi, H. J. Simpson,
T.-H. Peng, ibid. 206, 409 (1979).
4. R. M. Rotty and G. Marland, Oak Ridge Assoc.
Univ. Rep. IEA-80-9(M) (1980).
5. W. C. Wang, Y. L. Yung, A. A. Lacis, T. Mo, J.
E. Hansen, Science 194, 685 (1976).
6. National Academy of Sciences, Carbon Dioxide
and Climate: A Scientific Assessment (Washing-
ton, D.C., 1979). This report relies heavily on
simulations made with two three-dimensional
climate models (7, 8) that include realistic global
geography, seasonal insolation variations, and a
70-m mixed-layer ocean with heat capacity but
no horizontal transport of heat.
7. S. Manabe and R. J. Stouffer, Nature (London)
282, 491 (1979); J. Geophys. Res. 85, 5529
(1980).
8. J. Hansen, A. Lacis, D. Rind, G. Russell, P.
Stone, in preparation. Results of an initial CO2
experiment with this model are summarized in
(6).
9. National Academy of Sciences, Understanding
Climate Change (Washington, D.C., 1975).
10. R. E. Newell and T. G. Dopplick, J. Appl.
Meteorol. 18, 822 (1979).
11. S. B. Idso, Science 207, 1462 (1980); ibid. 210, 7
(1980).
12. J. Geophys. Res. 82 (No. 28) (1977); ibid. 85
(No. A13) (1980).
13. S. Manabe and R. T. Wetherald, J. Atmos. Sci.
24, 241 (1967).
14. A. Lacis, W. Wang, J. Hansen, NASA Weather
and Climate Science Review (NASA Goddard
Space Flight Center, Greenbelt, Md., 1979).
15. R. A. McClatchey et al., U.S. Air Force Cam-
bridge Res. Lab. Tech. Rep. TR-73-0096 (1973).
16. R. E. Roberts, J. E. A. Selby, L. M. Biberman,
Appl. Opt. 15, 2085 (1976).
17. 0. B. Toon and J. B. Pollack, J. Appl. Meteorol.
12, 225 (1976)._
18. R. D. Cess, J. Quant. Spectrosc. Radiat. Trans-
fer 14, 861 (1974).
19. W. C. Wang and P. H. Stone, J. Atmos. Sci. 37,
545 (1980).
20. R. D. Cess, ibid. 35, 1765 (1978).
21. G. Garrett, Dyn. Atmos. Oceans 3, 239 (1979);
P. Muller, ibid., p. 267.
22. S. L. Thompson and S. H. Schneider, J.
Geophys. Res. 84, 2401 (1979).
23. M. I. Hoffert, A. J. Callegari, C. T. Hsieh, ibid.
85, 6667 (1980).
24. H. Oeschger, U. Siegenthaler, U. Schotterer, A.
Gugelmann, Tellus 27, 168 (1975).
25. W. S. Broecker, Science 189, 460 (1975).
26. J. Hansen, A. Lacis, P. Lee, W. Wang, Ann.
N.Y. Acad. Sci. 338, 575 (1980).
27. H. H. Lamb, Philos. Trans. R. Soc. London
Ser. A 255, 425 (1970).
28. S. H. Schneider and C. Mass, Science 190, 741
(1975).
29. J. B. Pollack, 0. B. Toon, C. Sagan, A. Sum-
mers, B. Baldwin, W. Van Camp, J. Geophys.
Res. 81, 1971 (1976).
30. A. Robock, J. Atmos. Sci. 35, 1111 (1978);
Science 206, 1402 (1979).
31. W. Cobb, J. Atmos. Sci. 30, 101 (1973); R.
Roosen, R. Angione, C. Klemcke, Bull. Am.
Meteorol. Soc. 54, 307 (1979).
32. C. Sagan, 0. B. Toon, J. B. Pollack, Science
206, 1363 (1979).
33. S. Manabe and R. F. Strickler, J. Atmos. Sci.
21, 361 (1964).
34. V. Ramanathan, Science 190, 50 (1975).
35. B. G. Mendonca, Geophys. Monit. Clim.
Change 7 (1979).
36. R. Weiss, J. Geophys. Res., in press.
37. R. A. Rasmusson and M. A. K. Khalil, Atmos.
Environ. 15, 883 (1981).
38. F. S. Rowland and M. J. Molina, Rev. Geophys.
Space Phys. 13, 1 (1975).
39. R. L. Jenne, Data Sets for Meteorologial Re-
search (NCAR-TN/IA-I 11, National Center for
Atmospheric Research, Boulder, Colo., 1975);
Monthly Climate Data for the World (National
Oceanic and Atmospheric Administration,
Asheville, N.C.).
40. T. P. Barnett, Mon. Weather Rev. 106, 1353
(1978).
41. S. K. Kao and J. F. Sagendorf, Tellus 22, 172
(1970).
42. R. A. Madden, Mon. Weather Rev. 105, 9
(1977).
43. W. A. R. Brinkman, Quart. Res. (N. Y.) 6, 335
(1976); I. I. Borzenkova, K. Ya. Vinnikov, L. P.
Spirina, D. I. Stekhnovskiy, Meteorol. Gidrol.
7, 27 (1976); P. D. Jones and T. M. L. Wigley,
Clim. Monit. 9, 43 (1980).
44. P. E. Damon and S. M. Kunen, Science 193, 447
(1976).
45. J. E. Hansen, W.-C. Wang, A. A. Lacis, ibid.
199, 1065 (1978).
46. R. Oliver, J. Appl. Meteorol. 15, 933 (1976); C.
Mass and S. Schneider, J. Atmos. Sci. 34, 1995
(1977).
47. J. M. Mitchell, in Global Effects of Environmen-
tal Pollution, S. F. Singer, Ed. (Reidel, Dor-
drecht, Netherlands, 1970), p. 139.
48. D. V. Hoyt, Clim. Change 2, 79 (1979); Nature
(London) 282, 388 (1979).
49. K. Ya. Kondratyev and G. A. Nikolsky, Q. J.
R. Meteorol. Soc. 96, 509 (1970).
50. R. A. Madden and V. Ramanathan [Science 209,
763 (1980)] make a similar comparison for the
60'N latitude belt. It is generally more difficult
to extract the signal due to a global perturbation
from a geographically limited area.
51. W. W. Kellogg and R. Schware, Climate
Change and Society (Westview, Boulder, Colo.,
1978).
52. CLIMAP Project Members, Science 191, 1131
(1976).
53. H. H. Lamb, Climate: Present, Past and Future
(Methuen, London, 1977), vol. 2.
54. W. W. Kellogg, in Climate Change, J. Gribbin,
Ed. (Cambridge Univ. Press, Cambridge, 1977),
p. 205; Annu. Rev. Earth Planet. Sci. 7, 63
(1979).
55. J. J. Mercer, Nature (London) 271, 321 (1978);
T. Hughes, Rev. Geophys. Space Phys. 15, 1
(1977).
56. C. L. Parkinson and W. W. Kellogg, Clim.
Change 2, 149 (1979).
57. R. C. Willson, S. Gulkis, M. Janssen, H. S.
Hudson, G. A. Chapman, Science 211, 700
(1981).
58. S. H. Schneider and S. L. Thompson, J.
Geophys. Res., in press.
59. We thank J. Charney, R. Dickinson, W. Donn,
D. Hoyt, H. Landsberg, M. McElroy, L. Orn-
stein, P. Stone, N. Untersteiner, and R. Weiss
for helpful comments; I. Shifrin for several
typings of the manuscript; and L. DelValle for
drafting the figures.
SCIENCE, VOL. 213